UIT-Polar at SemEval-2026 Task 9 Detecting Multilingual, Multicultural and Multievent Online Polarization

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Advanced, short

Summary

The UIT-Polar system, a two-stage hybrid approach, addresses SemEval-2026 Task 9 for detecting multilingual and multievent online polarization. The initial stage utilizes DeBERTa for high-recall binary filtering, specifically designed to mitigate severe class imbalance in the dataset. Following this, the second stage employs Mistral for fine-grained polarization classification, which enhances semantic reasoning over the identified candidate instances. This coarse-to-fine architectural design improves overall system robustness and efficiency while crucially maintaining performance for minority classes. The system achieved Top-5 results on the English test set, demonstrating the effectiveness of combining encoder-based screening with large language model (LLM)-based refinement for this complex task.

Key takeaway

For Machine Learning Engineers developing online polarization detection systems, consider a two-stage hybrid architecture. Your initial stage should use a model like DeBERTa for high-recall binary filtering to manage severe class imbalance effectively. Subsequently, employ a large language model such as Mistral for fine-grained classification, enhancing semantic reasoning. This coarse-to-fine design improves robustness, efficiency, and crucial minority-class performance, offering a proven strategy for complex, imbalanced text classification.

Key insights

A two-stage hybrid system effectively detects online polarization by combining encoder-based screening with LLM-based refinement.

Principles

• Coarse-to-fine design improves robustness and efficiency.
• Prioritize minority-class performance in imbalanced data.
• Combine encoder and LLM for complex classification.

Method

A two-stage process: first, DeBERTa performs high-recall binary filtering to address class imbalance; second, Mistral conducts fine-grained polarization classification for semantic reasoning.

In practice

• Apply DeBERTa for initial high-recall filtering.
• Use Mistral for subsequent fine-grained classification.

Topics

• Online Polarization Detection
• Multilingual NLP
• Hybrid AI Systems
• DeBERTa
• Mistral
• Class Imbalance Mitigation

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• mdok-style at SemEval-2026 Task 9: Finetuning LLMs for Multilingual Polarization Detection — Finetuning mid-sized LLMs with QLoRA and augmented data improves multilingual online polarization detection.
• IHUBERT: Vector-Based Semantic Deduplication and Domain-Balanced Pretraining for Persian Resources — Semantic deduplication and domain-balanced pretraining significantly enhance Persian PLM performance across diverse NLU tasks.
• From Logistic Regression to GPT-2: Building a Complete Spam Detection & Sentiment Analysis Pipeline — For imbalanced text classification, prioritize F1-score and Precision-Recall AUC over accuracy and ROC-AUC.
• Comparing LLM and Fine-Tuned Model Performance on NVDRS Circumstance Extraction with Varying Prompt Complexity — LLMs, guided by a context-aware "Complexity Score" for prompt selection, significantly outperform fine-tuned models on inferentially complex, low-prevalence data extraction.
• JabuticaBERT: Modern Portuguese Encoders from Scratch with RTD and Long-Context Training — Monolingual encoders with careful architecture and tokenization can achieve competitive performance without massive scaling.
• Indigenous Letters to Brazil: Multi-Label Classification — Multi-label classification of indigenous letters presents challenges due to corpus imbalance and semantic overlap.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.